Professional Documents
Culture Documents
AmitKumarRoy80094100319 8 PDF
AmitKumarRoy80094100319 8 PDF
net/publication/264555752
CITATIONS READS
0 21,011
2 authors, including:
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Amit Kumar Roy on 10 August 2014.
Master of Engineering
in
Power Systems & Electric Drives
By:
Amit Kumar Roy
(Regn. No. 800941003)
JULY 2011
ELECTRICAL & INSTRUMENTATION ENGINEERING DEPARTMENT
THAPAR UNIVERSITY
PATIALA-147004
i
ACKNOWLEDGEMENT
I would like to thank Dr. Sanjay K. Jain, Associate. Prof., EIED for the opportunity to
work with him, for his invaluable guidance, encouragement, suggestions and untiring
support. He has been an advisor in true sense both academically and morally though out
this thesis work.
I am highly grateful to Thapar University, Patiala for providing all the necessary
infrastructure to complete my M.E. Thesis work.
I extend my sincere thanks to Dr. S. Ghosh, Professor & Head, EIED and to Dr.
Yaduvir Singh, Associate. Prof., EIED for their continuous inspiration during this thesis
work.
I am also thankful to all other faculty members of EIED, Thapar University, Patiala for
their intellectual support. Special thanks to my family members especially to my parents
and friends whose uninterrupted love, inspiration and blessings helped me to complete my
thesis study.
I thank and owe my deepest regards to all who have helped me directly or indirectly.
ii
ABSTRACT
Maintaining power system security is one of the challenging tasks for the power
system engineers. The security assessment is an essential task as it gives the knowledge
about the system state in the event of a contingency. Contingency analysis technique is
being widely used to predict the effect of outages like failures of equipment, transmission
line etc, and to take necessary actions to keep the power system secure and reliable. The
off line analysis to predict the effect of individual contingency is a tedious task as a power
system contains large number of components. Practically, only selected contingencies will
lead to severe conditions in power system. The process of identifying these severe
contingencies is referred as contingency selection and this can be done by calculating
performance indices for each contingencies.
The main motivation of the work is to carry out the contingency selection by
calculating the two kinds of performance indices; active performance index (PIP) and
reactive power performance index (PIV) for single transmission line outage. With the help
of Fast Decoupled Load Flow (FDLF), the PIP and PIV have been calculated in MATLAB
environment and contingency ranking is made. Further the contingency selection has been
done by using Radial Basis Function (RBF) Neural Network. This provides an effective
mean to rank the contingencies for various loading and generation levels in a power
system. The effectiveness of the method has been tested on 5-Bus, IEEE-14 Bus and
IEEE-30 Bus test systems.
iii
TABLE OF CONTENTS
Certificate i
Acknowledgement ii
Abstract iii
Table of Contents iv
List of Tables ix
1 INTRODUCTION……………………………………………………………...1
1.1 OVERVIEW……………………………....................................................................1
2.1 INTRODUCTION..................................................................................................8
iv
2.5 ALGORITHM FOR CONTINGENCY ANALYSIS USING FDLF…………........20
NEURAL NETWORK
3.1 INTRODUCTION……………………………………...........................................42
USING RBF-ANN…………………………………………………………..52
v
4 CONCLUSION AND FUTURE SCOPE……………………………………63
4.1 CONCLUSIONS………………………………………………………………..63
REFERENCES...………………………………………………………………………...64
APPENDIX-A....................................................................................................................68
APPENDIX-B....................................................................................................................69
APPENDIX-C...………………………………………………………………………….71
vi
LIST OF FIGURES
vii
Figure 2.18: CONTINGENCY RANKING AND PIV OF 30-BUS SYSTEM….…………...36
NEURAL NETWORK........................................................................................48
viii
LIST OF TABLES
TABLE 2.2: BUS VOLTAGES IN THE PRE AND POST CONTINGENCY STATE……25
TABLE 2.5: BUS VOLTAGES IN THE PRE AND POST CONTINGENCY STATE……31
TABLE 2.8: BUS VOLTAGES IN THE PRE AND POST CONTINGENCY STATE……37
ix
TABLE 3.3: REACTIVE POWER PERFORMANCE INDEX USING
x
CHAPTER-1
INTRODUCTION
1.1 OVERVIEW
Power system security involves system monitoring where the real time parameters
of the system are monitored by using the telemetry systems or by the SCADA systems. It
then involves the most important function of contingency analysis where the simulation is
being carried out on the list of “credible” outage cases so as to give the operators an
indication of what might happen to the power system in an event of unscheduled
equipment outage. This analysis forewarns the system operator, and allows deciding some
remedial action before the outage event.
For a power system to be secure, it must have continuity in supply without a loss of
load. For this security analysis is performed to develop various control strategies to
guarantee the avoidance and survival of emergency conditions and to operate the system at
lowest cost. Whenever the pre specified operating limits of the power system gets violated
the system is said to be in emergency condition. These violations of the limits result from
contingencies occurring in the system. Thus, an important part of the security analysis
revolves around the power system to withstand the effect of contingencies. The system
security assessment process is carried out by calculating system operating limits in the
1
pre- contingency and post contingency operating states at an operation control centre or at
the Energy Management System (EMS) of the utility company. The contingency analysis
is time consuming as it involves the computation of complete AC load flow calculations
following every possible outage events like outages occurring at various generators and
transmission lines. This makes the list of various contingency cases very lengthy and the
process very tedious. In order to mitigate the above problem, automatic contingency
screening approach is being adopted which identifies and ranks only those outages which
actually causes the limit violation on power flow or voltages in the lines. The
contingencies are screened according to the severity index or performance index where a
higher value of these indices denotes a higher degree of severity.
Security analysis involves the power system to operate into four operating states [1]:
Optimal dispatch: In this state the power system is in prior to any contingency. It
is optimal with respect to economic operation, but it may not be secure.
Post contingency: It is the state of the power system after a contingency has
occurred, it is being assumed that this condition has a security violation such as
line or transformer are beyond its flow limit, or a bus voltage is outside the limit.
Secure dispatch: It is the state of the system with no contingency, but with
corrections to the operating parameters to account for security violations.
Secure post-contingency: This is the state where the contingency is applied to the
base operating condition with corrections.
The concept of security analysis has been illustrated with a following example.
Suppose a power system consisting of two generators, a load, and a double circuit line, is
to be operated with both generators supplying the load as shown in Fig. 1.1(a) and
ignoring the losses it is assumed that the system as shown is in economic dispatch i.e. 500
MW is allotted for unit 1 and the 700 MW for unit 2 as the optimum dispatch. Further, it is
asserted that each circuit of the double circuit line can carry a maximum of 400 MW, so
that there is no loading problem in the base-operating condition. This condition is being
referred to as the optimal dispatch.
2
700 MW
500 MW 250 MW
Unit 2
Unit 1 250 MW
1200 MW
700 MW
500 MW 0 MW
Unit 2
Unit 1 500 MW
(OVERLOAD)
1200 MW
800 MW
400 MW 200 MW
Unit 2
Unit 1
200 MW
1200 MW
800 MW
400 MW 0 MW
Unit 2
Unit 1
400 MW
1200 MW
Now, a failure in one of the two transmission lines has been postulated and it can
be said that a line contingency has occurred and this results in change in power flows in
3
the other line causing the transmission line limit to get violated. The resulting flows have
been shown Fig. 1.1(b), this sate of power system is being said to be post contingency
state.
Thus by adjusting the generation on unit 1 and unit 2, the overloading in other line
is prevented and thus the power system remains secure. These adjustments are called
“security corrections.” Programs which can make control adjustments to the base or pre-
contingency operation to prevent violations in the post-contingency conditions are called
“security-constrained optimal power flows”. These programs can take account of many
contingencies and calculate adjustments to generator MW, generator voltages, transformer
taps etc. Together with the function of system monitoring, contingency analysis and the
corrective actions the analysis procedure forms a set of complex tools that can lead to the
secure operation of a power system.
5
studied in [25], to speed up the ANN training and performance. The use of Hopfield
neural network for contingency analysis has been shown in [26].
The use of Radial Basis Function (RBF) neural network for function
approximation proves to be highly efficient, as it employs a hybrid two stage learning
scheme. The RBF neural networks have several advantages like fast training, structural
simplicity and no local minima problem [27-28]. The non linear mapping capability of the
RBF network for estimating the line loading and bus voltage following a contingency in
bulk power systems has been done in [29]. Chicco et al. [30] presented a detailed
comparison for estimating the performance of self organising network, progressive
learning network and RBF. The author [31] suggested the parallel operation of RBF neural
network for contingency analysis to yield fast training and higher accuracy.
Contingency analysis problem has been solved by using fuzzy logic in [32]. The
post contingency cases have been first studied using fast decoupled load flow. These
quantities have been assigned a degree of severity. Fuzzy logic is applied to contingency
selection problem in [33] for voltage ranking where the post contingent voltages are used
to rank the contingencies. The application of Genetic algorithm for contingency ranking
has been studied in [34] where the problem of contingency ranking problem is formulated
as an optimization problem with an objective of finding the critical cases, this approach
reduces the computational burden for contingency analysis. The authors [35] attempted
post voltage calculations for a transmission line and transformer outages using Genetic
algorithm.
6
1.4 ORGANIZATION OF THESIS
The work carried out in this thesis has been summarized in four chapters, Chapter
1 deliberates on the overview of the problem, brief literature review, objectives of work
and organization of the thesis. Chapter 2 discusses the various methods for contingency
analysis, the contingency selection by calculating the active and reactive power
performance indices using the Fast decoupled load flow and the results for various
systems. In Chapter 3 the contingency analysis has been modelled using Radial basis
function neural network and corresponding results for various test bus systems. The
conclusions and the scope of further work are detailed in Chapter 4.
7
CHAPTER-2
8
done which involves the necessary security actions or necessary control to function in
order to mitigate the effect of contingency.
The generation shift factors are designated ali and have the following definition:
fl
ali
Pi (2.1)
where
l= line index
i=bus index
∆fl = change in megawatt power flow on line l when a change in generation ∆Pi occurs at bus i
∆Pi = change in generation at bus i
It is assumed that the change in generation ∆Pi is exactly compensated by an
opposite change in generation at the reference bus, and that all other generators remain
fixed. The ali factor then represents the sensitivity of the flow on line l due to a change in
generation at bus i. If the generator was generating Pio MW and it was lost, it is
represented by ∆Pi, as the new
o
∆Pi = - Pi (2.2)
power flow on each line in the network could be calculated using a pre calculated set of
“a” factors as follows:
fl = fl0 + ali ∆Pi for l = 1 . . . L (2.3)
where,
f l = flow on line l after the generator on bus i fails
f l0 = flow before the failure
9
The outage flow fl on each line can be compared to its limit and those exceeding
their limit are flagged for alarming. This would tell the operations personal that the loss of
the generator on bus i would result in an overload on line l. The generation shift sensitivity
factors are linear estimates of the change in flow with a change in power at a bus.
Therefore, the effects of simultaneous changes on several generating buses can be
calculated using superposition.
The line outage distribution factors are used in a similar manner, only they apply
to the testing for overloads when transmission circuits are lost. By definition, the line
outage distribution factor has the following meaning:
∆𝑓 𝑙
𝑑𝑙,𝑘 = (2.4)
𝑓𝑘0
where
dl,k = line outage distribution factor when monitoring line l after an outage on line k
∆fl = change in MW flow on line l
fk0 = original flow on line k before it was outaged i.e., opened
If one knows the power on line l and line k, the flow on line l with line k out can be
determined using "d" factors.
fl = fl0 + dl,k fk0 (2.5)
where
fl0 and fk0= pre outage flows on lines l and k, respectively
fl = flow on line l with line k out
By pre calculating the line outage distribution factors, a very fast procedure can be
set up to test all lines in the network for overload for the outage of a particular line.
Furthermore, this procedure can be repeated for the outage of each line in turn, with
overloads reported to the operations personnel in the form of alarm messages. The
generator and line outage procedures can be used to program a digital computer to execute
a contingency analysis study of the power system. It is to be noted that a line flow can be
positive or negative so that we must check fl against – flmax as well as flmax. It is assumed
that the generator output for each of the generators in the system is available and that the
line flow for each transmission line in the network is also available and the sensitivity
factors have been calculated and stored.
10
2.2.2 Contingency Analysis using AC Power Flow
The calculations made with the help of network sensitivity factors for contingency
analysis are faster, but there are many power systems where voltage magnitudes are the
critical factor in assessing contingencies. The method gives rapid analysis of the MW
flows in the system, but cannot give information about MVAR flows and bus voltages. In
systems where VAR flows predominate, such as underground cables, an analysis of only
the MW flows will not be adequate to indicate overloads. Hence the method of
contingency analysis using AC power flow is preferred as it gives the information about
MVAR flows and bus voltages in the system. When AC power flow is to be used to study
each contingency case, the speed of solution for estimating the MW and MVAR flows for
the contingency cases are important, if the solution of post contingency state comes late,
the purpose of contingency analysis fails. The method using AC power flow will
determine the overloads and voltage limit violations accurately. It does suffer a drawback,
that the time such a program takes to execute might be too long. If the list of outages has
several thousand entries, then the total time to test for all of the outages can be too long.
However, the AC power flow program for contingency analysis by the Fast Decoupled
Power Flow (FDLF) [9] provides a fast solution to the contingency analysis since it has
the advantage of matrix alteration formula that can be incorporated and can be used to
simulate the problem of contingencies involving transmission line outages without re
inverting the system Jacobian matrix for all iterations. Hence to model the contingency
analysis problem the AC power flow method, using FDLF method has been extensively
chosen.
11
Indices (PI) [1]. These indices are calculated using the conventional power flow
algorithms for individual contingencies in an off line mode. Based on the values obtained
the contingencies are ranked in a manner where the highest value of PI is ranked first. The
analysis is then done starting from the contingency that is ranked one and is continued till
no severe contingencies are found.
There are two kind of performance index which are of great use, these are active
power performance index (PIP) and reactive power performance index (PIV). PIP
reflects the violation of line active power flow and is given by eq.2.6.
L Pi
PIP = i=1(P )2n (2.6)
imax
where,
Pi = Active Power flow in line i,
Pimax = Maximum active power flow in line i,
n is the specified exponent,
L is the total number of transmission lines in the system.
If n is a large number, the PI will be a small number if all flows are within limit, and it
will be large if one or more lines are overloaded. Here the value of n has been kept unity.
The value of maximum power flow in each line is calculated using the formula
V i ∗V j
Pimax = (2.7)
𝑋
Another performance index parameter which is used is reactive power performance index
corresponding to bus voltage magnitude violations. It mathematically given by eq.2.8
2
𝑁𝑝𝑞 2(𝑉𝑖−𝑉𝑖𝑛𝑜𝑚 )
PIV= 𝑖=1 𝑉𝑖𝑚𝑎𝑥 −𝑉𝑖𝑚𝑖𝑛 (2.8)
12
Vinom is average of Vimax and Vimin
Npq is total number of load buses in the system
A flow chart for contingency selection technique is shown in Fig. 2.1
Start
Compute PIP
&
PIV
Increment the NO
Last outage done?
counter
YES
Stop
For calculation of PIV it is required to know the maximum and minimum voltage
limits, generally a margin of + 5% is kept for assigning the limits i.e, 1.05 P.U. for
13
maximum and 0.95 P.U. for minimum. It is to be noted that the above performance indices
is useful for performing the contingency selection for line contingencies only. To obtain
the value of PI for each contingency the lines in the bus system are being numbered as per
convenience, then a particular transmission line at a time is simulated for outage condition
and the individual power flows and the bus voltages are being calculated with the help of
fast decoupled load flow solution.
Vi yi1 V1
yi2 V2
Ii
Vn
yin
yi0
𝑛
Ii= 𝑗 =1 𝑌𝑖𝑗 𝑉𝑗 (2.9)
The above equation is expressed in polar form as
𝑛
Ii= 𝑗 =1 𝑌𝑖𝑗 𝑉𝑗 ∠𝜃𝑖𝑗 + 𝛿𝑗 (2.10)
The complex power at bus i is given by eq.2.11
Pi-jQi = Vi*Ii (2.11)
Substituting the value of (2.10) to (2.11) we get,
14
𝑛
Pi-jQi= 𝑉𝑖 ∠ − 𝛿𝑖 𝑗 =1 𝑌𝑖𝑗 𝑉𝑗 ∠𝜃𝑖𝑗 + 𝛿𝑗 (2.12)
Separating the real and real parts we get,
𝑛
Pi= 𝑗 =1 𝑌𝑖𝑗 𝑉𝑖 𝑉𝑗 cos 𝜃𝑖𝑗 − 𝛿𝑖 + 𝛿𝑗 (2.13)
𝑛
Qi=- 𝑗 =1 𝑌𝑖𝑗 𝑉𝑖 𝑉𝑗 sin 𝜃𝑖𝑗 − 𝛿𝑖 + 𝛿𝑗 (2.14)
Equations 2.13 and 2.14 are the non linear algebraic equations in terms of independent
variables, voltage and phase angle, we have two equations for load bus given by eq.2.13
and eq.2.14 and one equation for voltage controlled bus given by eq.2.14. Expanding
eq.2.13 and eq.2.14 in Taylor‟s series about the initial estimates we get the following set
of linear equations.
𝑘
𝑘 ∆𝛿2
∆𝑃2 𝜕𝑃2
𝑘
𝜕𝑃2
𝑘
𝜕𝑃2
𝑘
𝜕𝑃2
𝑘
… … .
. 𝜕𝛿 2𝑘 𝜕𝛿 𝑛𝑘 𝜕 𝑉2 𝜕 𝑉𝑛
.. .. .. .. .
. . . . .
. 𝜕𝑃𝑛
𝑘
𝜕𝑃𝑛
𝑘
𝜕𝑃2
𝑘
𝜕𝑃2
𝑘 .
∆𝑃𝑛 𝑘 … . .. ∆𝛿𝑛𝑘
𝜕𝛿 2𝑘 𝜕𝛿 𝑛𝑘 𝜕 𝑉2 𝜕 𝑉𝑛
_____ = _____ (2.15)
𝑘 𝑘 𝑘 𝑘
∆𝑄2𝑘 𝜕𝑄2
…
𝜕𝑄2 𝜕𝑄2
…
𝜕𝑄2
∆
(𝑘)
𝑉2
𝜕𝛿 2𝑘 𝜕𝛿 𝑛𝑘. 𝜕 𝑉2 𝜕 𝑉2
. .. ..
.. .. .
. . . .
𝜕𝑄2
𝑘
𝜕𝑄𝑛
𝑘
𝜕𝑄𝑛
𝑘
𝜕𝑄𝑛
𝑘 .
. … …
𝜕𝛿 𝑛𝑘 𝜕𝛿 𝑛𝑘 .
∆𝑄𝑛𝑘 𝜕 𝑉2 𝜕 𝑉𝑛
∆ 𝑉(𝑘)
𝑛
From eq.2.15 the Jacobian matrix gives the linearized relationship between small
(𝑘) (𝑘)
changes in voltage angle ∆ δ𝑖 and voltage magnitude ∆ 𝑉𝑖 with the small changes in
real and reactive power ∆𝑃𝑖 𝑘 and ∆𝑄𝑖 𝑘 . The elements of Jacobian matrix are the partial
(𝑘) (𝑘)
derivatives of equations 2.13 and 2.14 evaluated at ∆ δ𝑖 and ∆ 𝑉𝑖 . In short the eq.2.15
Using these, the new values of active and reactive power are being found using equations
2.13 and 2.14, simultaneously the new values of the elements of Jacobian matrix are being
obtained using equations 2.17 – 2.24 and again the new values of phase angles and bus
(𝑘) (𝑘)
voltages are obtained and the process is continued until the residuals ∆𝑃𝑖 and ∆𝑄𝑖 are
less than he specified accuracy, i.e.,
(𝑘)
∆𝑃𝑖 ≤ɛ (2.29)
(𝑘)
∆𝑄𝑖 ≤ɛ (2.30)
In the above process the calculation of new values of phase angles and bus voltages are
obtained only after the inversion of the Jacobian matrix which is very complex and
requires rigorous computational effort so as to avoid this the concept of FLDF is used.
16
It has been found that for transmission lines have high values of X/R ratio, so for such
system the real power changes ∆P are less sensitive to changes in the voltage magnitude
and are most sensitive to change in phase angles ∆𝜹. Similarly the reactive power is less
sensitive to change in angles and is mainly dependent on changes in voltage magnitude;
this concept is used to formulate the FDLF solution. Thus the element J2 and J3 of the
Jacobian matrix is set to zero. Thus eq.2.16 can be written as
∆𝑃 𝐽 0 ∆𝛿
= 1 (2.31)
∆𝑄 0 𝐽4 ∆𝑉
or
𝜕𝑃 𝑖
∆𝑃 = 𝐽1 ∆𝛿 = ∆𝛿 (2.32)
𝜕𝛿 𝑗
𝜕𝑃 𝑖
∆𝑄 = 𝐽4 ∆ 𝑉 = ∆𝑉 (2.33)
𝜕𝑉
Hence we can say that the matrix equations 2.32 and 2.33 have separated into two
decoupled equations requiring considerably less amount of time as compared to eq.2.16,
further the recomputing of J1 and J4 per iteration gets eliminated. The diagonal elements of
J1 described by eq.2.17 can be written as
𝜕𝑃 𝑖 𝑛
= 𝑗 =𝑖 𝑌𝑖𝑗 𝑉𝑖 𝑉𝑗 sin 𝜃𝑖𝑗 − 𝛿𝑖 + 𝛿𝑗 - 𝑉𝑖2 𝑌𝑖𝑖 sin 𝜃𝑖𝑖 (2.34)
𝜕𝛿 𝑖
Replacing the first term of the above equation with -𝑄𝑖 as given by (2.14) results in
𝜕𝑃 𝑖
= - 𝑄𝑖 - 𝑉𝑖2 𝑌𝑖𝑖 sin 𝜃𝑖𝑖 (2.35)
𝜕𝛿 𝑖
17
Further simplification is obtained by assuming 𝑉𝑗 ≈ 1
𝜕 𝑃𝑖
= - 𝑉𝑖 𝐵ij (2.39)
𝜕𝛿 𝑗
Replacing the second term of the above term of the above equation with -𝑄𝑖 as given by
eq.2.14 results in
𝑘
𝜕𝑄𝑖
= - 𝑌𝑖𝑖 𝑉𝑖 sin 𝜃𝑖𝑖 + 𝑄𝑖 (2.41)
𝜕 𝑉𝑖
Again, since 𝐵𝑖𝑖 = 𝑌𝑖𝑖 sin 𝜃𝑖𝑖 >> 𝑄𝑖 , 𝑄𝑖 may be neglected and eq.2.23 reduces to
𝑘
𝜕𝑄𝑖
= - 𝑉𝑖 𝐵𝑖𝑖 (2.42)
𝜕 𝑉𝑖
∆𝑃
= −𝐵′ ∆𝛿 (2.44)
𝑉𝑖
∆𝑄
𝑉𝑖
= −𝐵′′ ∆ 𝑉𝑖 (2.45)
Here, 𝐵 ′ and 𝐵 ′′ are the imaginary part of the bus admittance matrix YBUS, therefore in
fast decoupled load flow power algorithm, the successive voltage magnitude and phase
angle changes are given by,
∆𝑃
∆𝛿 = -[𝐵′ ]-1 (2.46)
𝑉
∆𝑄
∆ 𝑉 = -[𝐵′′ ]-1 (2.47)
𝑉
Thus it can be concluded that fast decoupled power flow solution requires the least time
per iteration among all load flow techniques available, hence the power flow solution is
obtained very rapidly. Thus this technique is very useful in contingency analysis where
numerous outages are to be simulated in a very rapid manner.
18
The flow chart for Fast Decoupled Load Flow solution has been shown in Fig. 2.3.
Start
P
Calculate
|V |
Yes
Does Q Yes
Increment counter Does P converge ?
converge ?
K=K+1 for next iteration
No No
Calculate
P
[ B ]1 Print
'
Stop
|V | Results
Q
Calculate | V |
Yes
Yes Does Q
Does P converge ?
converge ?
No
No
Q
'' 1
V [ B ]
|V |
19
2.5 ALGORITHM FOR CONTINGENCY ANALYSIS USING
FAST DECOUPLED LOAD FLOW
The algorithm steps for contingency analysis using fast decoupled load flow
solution are given as follows:
Step 1: Read the given system line data and bus data.
Step 4: Calculate the active power flow for in the remaining lines and the maximum
power flow PMax using eq.2.7.
Step 5: Calculate the active power performance index PIP which give the indication of
active power limit violation using eq.2.6.
Step 6: Calculate the voltages at all the load buses following the line contingency.
Step 7: Calculate the reactive power performance index PIV which gives the voltage limit
violation at all the load buses due to a line contingency using eq.2.8.
Step 8: Check if this is the last line outage to be simulated; if not the step (3) to (7) is
computed till last line of the bus system is reached.
Step 9: The contingencies are ranked once the whole above process is computed as per the
values of the performance indices obtained.
Step 10: Do the power flow analysis of the most severe contingency case and print the
results
20
Start
Last Yes
contingency
reached ?
No
Stop
5
54.82 MW 27.93 MW
(1.060 pu)
(1.024 pu)
88.86 MW (1.047 pu)
40.72 MW 24.69 MW
1 2
18.87 MW
3
(1.024 pu)
Table 2.1
Performance Indices & Contingency Ranking using FDLF for 5-Bus System
Outage Line No) PIP PIV Ranking
1 0.2800 3.1916 1
2 0.3619 0.2699 6
3 0.3377 0.6557 4
4 0.3790 0.6173 5
5 0.4221 0.2653 7
6 0.2995 0.8599 3
7 0.3036 0.8799 2
It is seen that the contingency in the line connected between buses (1-2) results in
highest value of the reactive power performance index and thus it is ranked first for the
contingency selection, hence the post contingency state of the system corresponding to
this contingency has been analysed. Since, the value of PIV indicates the severity that is
occurring in the system due to violation in the voltage limits, the pre contingency and the
post contingency voltages at the buses of the entire system have been detailed in Table
2.2. The MW flows corresponding to the pre contingency state and the post contingency
state have been detailed in Table 2.3.
23
0.45
0.4
0.35
0.3
0.25
PIp
0.2
0.15
0.1
0.05
0
1 2 3 4 5 6 7
Outage line Number
3.5
2.5
2
PIv
1.5
0.5
0
1 2 3 4 5 6 7
Outage line Number
24
4
3.5
2.5
PIv
1.5
0.5
0
1 2 3 4 5 6 7
Rank of Line outage Contingency
Table 2.2
Bus Voltages in the Pre and Post Contingency State
Bus Number Pre-contingency voltage Post-contingency voltage
(pu) (pu)
1 1.060 1.060
2 1.047 0.891
3 1.024 0.886
4 1.024 0.880
5 1.018 0.861
25
Table 2.3
Active Power Flow in the Pre and Post Contingency State
Line No Start Bus End Bus Pre contingency Post contingency
MW flow MW flow
1 1 2 88.86 MW 0 MW
2 1 3 40.72 MW 143.63 MW
3 2 3 24.69 MW 15.33 MW
4 2 4 27.93 MW 4.00 MW
5 2 5 54.82 MW 39.07 MW
6 3 4 18.87 MW 66.80 MW
7 4 5 6.33 MW 22.23 MW
Fig. 2.9 shows the power flows and bus voltages in the system following the most
severe contingency, which is the outage of the transmission line connecting buses (1-2).
5 4
22.23 MW
(0.861 pu)
39.07 MW 4.00 MW
1
(0.880 pu)
(0.891 pu)
0 MW
143.63 MW 15.33 MW
2
(1.060 pu) 66.8 MW
3 (0.886 pu)
26
corresponding to the base case loading condition is also shown in Fig. 2.10. This state of
the system corresponds to the pre contingency state. The system has a total 20 number of
transmission lines, hence we evaluate for 20 line contingency scenarios by considering the
one line outage contingency at a time. The performance indices are summarized in the
Table 2.4. From Table 2.4 it can be inferred that outage in line number 16 is the most
vulnerable one and its outage will result a great impact on the whole system. The high
value of PIV for this outage also suggests that the highest attention be given for this line
during the operation. Fig. 2.11 and Fig. 2.12 shows the graphical representation of the
performance indices for all the line contingencies with the value of PI on the y-axis and
the outage line number labelled on the x-axis.
G GENERATORS
SYNCHRONOUS
C
COMPENSATORS
13
6.14 MW
12
14
1.82 MW 18.2 MW
11
4.32 MW 8.94 MW
7.93 MW 10
7.96 MW
4.72 MW 9
C
G
7
6 15.8 MW 8
1 C
44.74 MW 4
74.7 MW 60.19 MW
5
157.3 MW
56.3 MW
42.03 MW
23.19 MW
2
71.6 MW
3
G
27
Table 2.4
Performance Indices & Contingency Ranking using FDLF for 14-Bus System
2 0.9807 7.6696 11
3 1.1654 10.0014 7
4 0.9999 7.3213 12
5 0.9820 8.8759 9
6 0.9640 13.2572 2
7 0.9915 0.3566 19
8 1.0747 1.1753 17
9 0.9807 10.5776 4
10 1.2396 1.6047 16
11 1.0142 9.5907 8
12 1.0127 1.8089 15
13 1.0569 1.3669 18
14 1.0072 10.4518 6
15 1.0759 0.0844 20
16 1.0114 13.3464 1
17 1.0164 2.3482 13
18 1.0030 10.5217 5
19 1.0008 12.5538 3
20 1.0076 2.2891 14
28
1.4
1.3
1.2
1.1
0.9
0.8
PIp
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Outage Line Number
14
12
10
8
PIv
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Outage Line Number
12
10
8
PIv
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Rank of Line outage Contingency
The contingencies have been ordered by their ranking where the most severe
contingency is being ranked 1 and the least has been ranked 20. The variation of reactive
performance index with their ranking has been shown in the Fig. 2.13. It is clear from the
result of different PIV that the contingency number 16 which the line outage contingency
corresponding to the line connected between buses (9-10) is the most severe
contingency. Hence the post contingency analysis corresponding to this line outage has
been performed. The voltage of the system corresponding to the pre contingency state and
the post contingency state has been detailed in Table 2.5. The MW flows corresponding to
the pre contingency state and the post contingency state has been detailed in Table 2.6.
The state of the system after the outage in the line connected between buses (9-10) has
been represented in Fig. 2.14.
30
Table 2.5
Bus Voltages in the Pre and Post Contingency State
Bus Number Pre-contingency voltage Post-contingency voltage
(pu) (pu)
1 1.075 1.075
2 1.050 1.050
3 1.000 1.000
4 1.002 1.000
5 1.009 1.009
6 1.025 1.025
7 1.007 1.008
8 1.016 1.016
9 0.993 0.996
10 0.991 0.978
11 1.004 0.997
12 1.007 1.008
13 1.001 1.002
14 0.978 0.980
Table 2.6
Active Power Flow in the Pre and Post Contingency State
Line No Start Bus End Bus Pre contingency Post contingency
MW flow MW flow
1 1 2 157.3 MW 156.8 MW
2 1 5 74.7 MW 75 MW
3 2 3 71.6 MW 71.5 MW
4 2 4 56.3 MW 56 MW
5 2 5 42.03 MW 42.3 MW
6 3 4 23.19 MW 23 MW
7 4 5 60.19 MW 57.6 MW
8 4 7 27.38 MW 25.5 MW
9 4 9 15.8 MW 14.9 MW
10 5 6 44.74 MW 48.2 MW
11 6 11 7.96 MW 12.79 MW
12 6 12 7.93 MW 7.58 MW
13 6 13 18.21 MW 16.67 MW
14 7 8 0.0 MW 0.0 MW
15 7 9 27.39 MW 25.5 MW
16 9 10 4.72 MW 0 MW
17 9 14 8.94 MW 10.7 MW
18 10 11 4.32 MW 9.09 MW
19 12 13 1.82 MW 1.4 MW
20 13 14 6.14 MW 4.33 MW
31
G GENERATORS
SYNCHRONOUS
C
COMPENSATORS
13
4.33 MW
12 16.67 MW 14
1.4 MW
11
9.09 MW
10 10.7 MW
7.58 MW
12.79 MW
0 MW 9
C
G
7
6 14.9 MW 8
1 C
48.2 MW 4
75 MW
57.6 MW
5
156.8 MW
56 MW
42.3 MW
23 MW
2
71.5 MW
3
G
32
1
177.77 MW G
2
G
83.22 MW
45.71
MW 14
3 13 1.55 MW
C
7.89 MW 15
4
5.005
17.82 MW MW 30
61.9 12
MW 6.02 MW
16 23
83 MW 70.12 MW 7.20 MW
24
18
3.65 MW 2.78 7.09 3.70
MW MW MW
6.74 1.33
20 MW 19
MW
9.02 MW
10 21
5 6
14.35 37.52 5.64
MW MW 17 15.73 MW MW
C 9 27
29
4.90
7.58 MW 22
MW
7 25
29.5
28
11
MW C
18.81 MW
3.53 MW
26
0.57 MW
8
33
The active and reactive power performance indices are also calculated considering
the outage of only one line sequentially and the calculated indices are detailed in Table
2.7. From Table 2.7 it can be inferred that outage of line number 9 is the most vulnerable
one and its outage will result a great impact on the whole system. The highest value of PIV
for his outage suggests that highest attention be given for this line during the operation.
The graphical representation of the PIP and PIV values for the corresponding line
contingencies respectively has been shown in Fig. 2.16 and Fig. 2.17 respectively.
Table 2.7
Performance Indices & Contingency Ranking using FDLF for 30-Bus System
2.5
1.5
PIp
0.5
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41
Outage Line Number
35
30
25
20
PIv
15
10
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41
Outage Line Number
30
25
20
PIv
15
10
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41
Rank of Line outage Contingency
36
Fig. 2.18 shows the contingency ranking of this system with respect to the PIV
values. Since for the IEEE-30 bus system contingency number 9 which is the line
connected between buses (6-7) is the most critical contingency, the post contingency
analysis following the outage of this line has been done. Here the system‟s bus voltages
corresponding to the pre contingency and the post contingency state has been obtained.
The results are detailed in Table 2.8 and the active power flow in all the lines in the pre
contingency and post contingency state has been detailed in Table 2.9.
Table 2.8
Bus Voltages in the Pre and Post Contingency State
Bus Number Pre-contingency voltage Post-contingency voltage
(pu) (pu)
1 1.060 1.060
2 1.043 1.043
3 1.022 1.024
4 1.013 1.016
5 1.010 1.010
6 1.012 1.015
7 1.003 0.988
8 1.010 1.010
9 1.051 1.053
10 1.044 1.047
11 1.082 1.082
12 1.057 1.059
13 1.071 1.071
14 1.042 1.044
15 1.038 1.039
16 1.045 1.046
17 1.039 1.041
18 1.028 1.030
19 1.025 1.027
20 1.029 1.031
37
21 1.032 1.034
22 1.033 1.035
23 1.027 1.029
24 1.022 1.024
25 1.019 1.021
26 1.001 1.004
27 1.026 1.028
28 1.011 1.013
29 1.006 1.009
30 0.995 0.997
Table 2.9
Active Power Flow in the Pre and Post Contingency State
Line No Start Bus End Bus Pre contingency Post contingency
MW flow MW flow
1 1 2 177.77 MW 187.77 MW
2 1 3 83.22 MW 74.80 MW
3 2 4 45.71 MW 32.14 MW
4 3 4 78.01 MW 70.13 MW
5 2 5 82.99 MW 123.95 MW
6 2 6 61.91 MW 43.81 MW
7 4 6 70.12 MW 50.83 MW
8 5 7 14.35 MW 23.06 MW
9 6 7 37.52 MW 0 MW
10 6 8 29.5 MW 29.55 MW
11 6 9 27.69 MW 28.42 MW
12 6 10 15.82 MW 16.25 MW
13 9 11 0.00 MW 0.00 MW
14 9 10 27.69 MW 28.42 MW
15 4 12 44.12 MW 42.66 MW
16 12 13 0.00 MW 0.00 MW
17 12 14 7.89 MW 7.69 MW
38
18 12 15 17.82 MW 17.23 MW
19 12 16 7.20 MW 6.53 MW
20 14 15 1.55 MW 1.43 MW
21 16 17 3.65 MW 2.98 MW
22 15 18 6.02 MW 5.65 MW
23 18 19 2.78 MW 2.41 MW
23 19 20 6.74 MW 7.10 MW
25 10 20 9.02 MW 9.39 MW
26 10 17 5.37 MW 6.03 MW
27 10 21 15.73 MW 15.81 MW
28 10 22 7.58 MW 7.63 MW
29 21 22 1.87 MW 1.79 MW
30 15 23 5.00 MW 4.59 MW
31 22 24 5.64 MW 5.71 MW
32 23 24 1.77 MW 1.36 MW
33 24 25 1.33 MW 1.60 MW
34 25 26 3.53 MW 3.54 MW
35 25 27 4.90 MW 5.17 MW
36 28 27 18.18 MW 18.45 MW
37 27 29 6.18 MW 6.18 MW
38 27 30 7.09 MW 7.08 MW
39 29 30 3.70 MW 3.70 MW
40 8 28 0.57 MW 0.55 MW
41 6 28 18.81 MW 19.07 MW
The state of the entire system after the outage in the transmission line connected
between buses (6-7) has been shown in Fig. 2.19.
39
1
187.71 MW G
2
G
74.80 MW
32.14
MW 14
3 13 1.43 MW
70.13
MW C
7.69 MW 15
4
4.5
17.23 MW MW 30
43.81 12
MW 5.65 MW
16 23
123.95 MW 50.83 MW 6.53 MW
18 24
2.98 MW 2.41 7.08 3.70
MW MW MW
7.10
20 MW 19
9.39 MW
10 21
5 0 1.6
23.06 6 MW
MW
MW 17 15.81 MW 27
C 9 5.71
MW 5.17 29
7.63 MW 22 MW
7 29.5 25
MW
28
11
C
19.07 MW
3.54 MW
0.55 MW 26
40
2.7 CONCLUDING REMARKS
The methods of contingency analysis using sensitivity factors and AC power flow
have been presented, the analysis with AC power flow using FDLF is found most suitable.
Since, the list of possible contingency cases is very large for a complex network like
power system, hence the approach of contingency selection plays a very important role as
it eliminates the large number of contingency cases and focuses on the most severe
contingency case. It is highly demanding that the entire process of contingency analysis is
done in least time. Hence, to speed up the contingency analysis process as a whole, the
computing speed in the selection process must be enhanced.
From the results obtained it can be concluded that the calculation of performance
indices gives a good measure about the severity of all the possible line contingencies
occurring in the system. The indices with highest value reflect a severe case which has the
highest potential to make the system parameters to go beyond their limits. Hence, the most
severe contingency case has been chosen from the list of various line contingencies and
the post contingency analysis pertaining to this contingency has been done where the most
important system parameters like bus voltages and the MW flows have been calculated.
The list of severity of contingencies before the power system is put to operation acts as a
useful guide to run a reliable system.
41
CHAPTER-3
3.1 INTRODUCTION
The contingency analysis process by conventional load flow solution using FDLF
can give the solution to the contingency selected only for one loading and generating
condition at one go. But practically a power system can have a varying level of operating
conditions and to predict the solution of contingency analysis for all the possible and
future operating points is an impossible task. In such cases, the use of Artificial Neural
Networks (ANN) can be useful. Since ANN has the ability to predict the output for unseen
input set once it gets trained with sufficient number of training patterns. It has been used
for dynamic nature problems in power system like contingency analysis, load forecasting,
component fault detection etc. The ability of Neural Networks to predict the outcome for a
new pattern in a fast and accurate manner makes them suitable for online analysis also.
This chapter discusses the brief review of ANN, its application to power system
and its ability to solve the contingency analysis problem. The two case studies are used to
investigate the solution of contingency selection problem using Radial Basis Function
Neural network and the results are discussed thoroughly.
42
The first model of ANN was proposed by McCulloch & Pitt and following it many
researchers proposed various other models of neural net like the Hebbs model, Rossenblatt
perceptron model, Hopfield neural network model, Radial Basis Function neural networks
and the Vector support machine network model is one of the latest model in the neural net
topology [27].
X1
W1
y
f(x)
W2
X2
The basic structure of ANN [27] is shown in Fig. 3.1 where the nodes signify the
processing units which are analogous to neurons of a brain, these nodes receives the
signals and are transmitted by means of connecting links, these links posses an associated
weights which are multiplied by the incoming signal and the output is obtained by
applying the activations to the net input.
Out of the various neural network topologies available, the neural networks
differed in terms of their architecture, learning or training process and activation
functions. In the following section the characterization of the neural networks based on
these parameters have been discussed in brief.
Various types of architecture feature are: feed forward network, feed back
network, fully recurrent network and competitive network [28].
Feed Forward Net: These networks may have single layer of weights where the inputs
are directly connected to the outputs or multiple layers with intervening sets of hidden
units. Hidden units are used to create internal representations of the input patterns and
with adequate number of hidden units, it is possible to approximate arbitrarily any
function with a simple feed forward network.
43
Competitive Net: These networks are similar to single layer feed forward networks
except that there are connections, usually negative, between the output nodes, because of
these connections the output nodes tend to compete to represent the current input pattern.
These outputs can be completely connected and some times the connections are restricted
to units that are close to each other. With an appropriate learning algorithm these networks
can be made to organize it topologically.
Recurrent Net: In this network architecture all the units are connected to all other units
and every unit act as both input and output. Typically, a set of patterns is instantiated on
all of the units, one at a time. As each pattern is instantiated the weights are modified,
when a degraded version of one of the patterns is presented, the network attempts to
reconstruct the pattern.
The process of setting the perfect value of weights between the network layers
with an objective to extract an exact output is called training a network. The internal
process that takes place during the training is referred to as learning. The common
training methodologies [28] have been discussed in brief:
Supervised Training: This kind of training can be considered to as training with the
help of a teacher where the new set of inputs are being continuously fed to the network
and the output generated from the network is compared with the set of previously
generated target data, the error between the actual and the target data is being calculated
and based upon this the learning rule is being employed to modify the synaptic weights
between the layers. Supervised training is widely used for pattern association; some
algorithms employing supervised training are Hebb‟s rule, delta learning rule, error back
propagation etc.
44
can be achieved. Since they have the ability to self learn the networks employing this
training paradigm is being referred to as self organizing networks.
The Activation functions [27] are used to calculate the output response of a
neuron, the sum of weighted input signal is being applied with an activation to obtain the
response, these functions can be linear as well as non linear, where the non linear
activation function are used in multilayer networks. Some of the commonly used
activation functions are shown in Fig. 3.2, these are detailed in brief.
F(x)
F(x)
x
X-----
F(x)
F(x)
2 1
1
1
-1
0 x
The Binary Step Function is given by eq.3.1 here 𝜽 is called the threshold of the
activation function the function has been shown in Fig. 3.2(a)
f(x) = 1; if f(x) ≥ 𝜽
= 0; if f(x) ≤ 𝜽 (3.1)
45
also known as radial basis function as it is used as an activation function in the hidden
neurons of the radial basis neural network. This function is illustrated in Fig. 3.2(b)
𝟐
| 𝑿𝒊−𝐭𝐣 |
h 𝑿𝒊 − 𝒕𝒋 = 𝒆𝒙𝒑 − 𝟐𝛔𝟐
(3.2)
where,
The Binary Sigmoidal Functions are usually S-shaped and it is also known as
hyperbolic functions or the logistic functions and these are usually used in multilayer
networks like the back propagation network. It ranges between 0 and 1 and it is given by;
1
𝑓 𝑥 = (3.3)
1+exp
(−𝜎𝑥 )
The Bipolar Sigmoidal Function ranges between +1 and -1, and the function is
related as hyperbolic tangent function. The bipolar sigmoidal function is given as;
1−exp
(−𝜎𝑥 )
𝑓 𝑥 = (3.4)
1+exp
(−𝜎𝑥 )
Mostly it is found that bipolar data is generally used hence this activation function is
widely used. Fig. 3.2(d) shows the bipolar sigmoidal function.
The Radial Basis Fuction (RBF) neural networks were first independently
proposed by Broomhead et al. [22]. This artificial network model is mainly motivated by
the “locally tuned” response observed in biological neurons. The RBF architecture has a
simple feed forward architecture consisting of three layers, an input layer, one hidden
layer and one output layer as shown in Fig. 3.3. The hidden layer activation in RBF is
computed using an exponential of a distance measure called the Euclidean distance norm,
this layer performs the non linear mapping of the inputs and it commonly uses the
Gaussian activation function denoted by Ω(x). Let the RBF network shown in Fig. 3.3
consists an n-dimensional input vector X and a total of J-neurons in the hidden layer
which are fully interconnected to an output layer of L-neurons. The outputs of the hidden
unit are not calculated using the weighted sum or sigmoidal activation mechanism as used
in other networks, rather they are calculated by the exponential measure between the input
pattern Xi and the bias centre vector tj. It is to be noticed that the choice of bias centres are
taken randomly from the input training data and the number of bias centres are equal to the
number of hidden neurons. The output of the hidden unit h 𝑿𝒊 − 𝒕𝒋 is given by eqn.3.5
𝑿𝒊 −𝒕𝒋 𝟐
h 𝑿𝒊 − 𝒕𝒋 = 𝒆𝒙𝒑 − 𝟐𝛔𝟐
(3.5)
where ,
47
i takes the value from 1 to n,
[W]
y1
X1
X2 y2
Xn yL
Weighted Output layer
Input layer
connections
Hidden layer
It is to be noted that the input is passed from the input layer to the hidden layer
with out any transformation. After computing the signal in accordance to the Gaussian
function, the activated output from the hidden neuron is passed on to the linear output
neurons through the weighted connected nodes. The size of the weight matrix is L x J and
the output from the activity vector y is given by eqn.3.6
48
𝑱
𝒚𝒍 = 𝒋=𝟏 𝑾𝒍𝒋 𝒉 𝑿𝒊 − 𝒕𝒋 (3.6)
where, Wlj is the synaptic weights connecting hidden neuron j to output neuron l and J is
the number of neuron in the hidden layer. There is no formal method for specifying the
required number of hidden units in an RBF network. It is to be noted that the weight of the
RBF network is determined by a combination of supervised and unsupervised learning, the
learning which takes place between the input and hidden layer is unsupervised one and the
learning that takes between the hidden and output layer is supervised learning. As the
network is required to fit the training data with the desired output dj , where j takes the
value from 1 to n. It is required to fit the training data as per the eqn.3.7
yl = d j (3.7)
𝑱
𝒅𝒋 = 𝒋=𝟏 𝑾𝒍𝒋 𝒉 𝑿𝒊 − 𝒕𝒋 (3.8)
H.W = d (3.9)
The required weight between the hidden and output unit can be found directly from the
above equation. The H matrix is not a square matrix, therefore no unique inverse exists for
the H matrix. Therefore to calculate the weight W, the minimum norm solution as
explained below is used;
W = H+d
= (HTH)-1HTd (3.10)
49
3.5 RBF APPLIED TO CONTINGENCY ANALYSIS
The two performance indices parameters PIP and PIV indicating a measure about
the severity of the line contingency as discussed in section 2.6, have been evaluated by the
use of RBF-ANN. It is clear that when any neural network is applied for a practical
problem two modes of operation namely training and testing are performed. In training,
sets of training data are used to adjust the weights of the network. Where as, in testing the
network response is being seen for the new data which had not been used in training. For
the contingency analysis problem the training data has been obtained by using
conventional load flow solution for different loading levels and generation scenarios. The
training phase has been carried out till the error between the desired and the actual output
is small. Where as, testing is done using the data of loading levels that had not been used
in the training phase. The RBF-ANN consists of three layers namely input, hidden and
output layer, it‟s the hidden layer which computes the output by an exponential measure
between the input data and a sample data. A generalized adopted model for the input,
hidden and output layer which has been used for Contingency analysis is shown in Fig.
3.4, the salient features are:
The Input layer should include as many neurons as required for the desired input
information. Generally the power injections for generator and load bus are chosen
as the raw inputs to the ANN. It is advantageous to choose the power injections as
the input data since they are readily available where as the parameters like bus
voltages and phase angles cannot be obtained directly. The input layer [x] consists
of power injections, P and Q at the generator and load buses, and the value Ki
which represents the outage of line i, a sample of input pattern has been shown:
[x] = [PG1, QG1,………PGg, QGg,
PL1, QL1,……….PLn, QLn,Ki] (3.11)
Where,
50
The Middle layer does not have any criteria for selecting the number of neurons
and the choice of the number of neurons in this layer is based on experimentation
and simulation, in general the more the number of neurons in the middle layer the
better the network can fit the targets while too many neurons in the middle layer
can result in over fitting hence the process of experimentation is purely followed
for selecting the number of neurons.
The Output layer consists of an output vector [O] with two elements which are the
active and reactive power performance indices PIP and PIV respectively, i.e.,
[O] = [PIP, PIV] (3.12)
QG1
Hidden layer
Output layer
PG2
PIp
QG2
PL1
PIv
QL1
K
Synaptic weights connecting hidden
and output layer
Input layer
51
3.5.1 Algorithm for PIP and PIV Indices prediction using RBF-ANN
The solution algorithm for the prediction of the active and reactive power
performance indices (PIP & PIV) is given below:
1. Carry out the fast decouple load flow analysis for all the single line outage cases
and calculate the PIP and PIV. The indices obtained from this stage are considered as
the desired output.
2. Select inputs for the RBFN [x] = [PG1,QG1,……..,PGg,QGg,
PL1,QL1,………,PLn,QLn,Ki]
3. Select the number the number of neurons in the middle layer and determine the bias
centres tj from the input training vector.
4. Select the appropriate value of the spread 𝛔.
5. Calculate the output of the jth hidden unit using the eq.3.5.
6. Calculate the output of the lth unit of the output neuron yl using eq.3.6.
7. Calculate the weight matrix that solves the network specification using eq.3.10.
8. Select the test data for the network, these data are chosen different from the data
that had been used for the training purpose.
9. Use the weight matrix obtained in step (7) to compute the output from the output
layer.
10. Check if the output obtained in step (9) is close to the desired output, if yes then
stop, else change the value of number of bias centres and spread and repeat steps
(5-9).
The flow chart of the above algorithm has been illustrated in the Fig. 3.5
52
START
Calculate weight
matrix using eq.3.10
53
3.6 RESULTS AND DISCUSSION
In this section, the results of contingency analysis problem using RBF neural
network have been presented. The algorithms are implemented in MATLAB for the
above. The main objective is to determine the active and reactive power performance
indices which form an important part of contingency analysis for different bus systems.
The algorithms have been evaluated on two set of bus systems which are being referred as
Case I and Case II.
Case I: 5 - Bus System [36].
Case II: 14 - Bus Test System [37].
3.6.1 Case I: 5- Bus System
The bus data and line data specifications of the five bus test system have been
given in APPENDIX A. The results of active power performance index PIP and reactive
power performance indices PIV for the base case loading condition of 1650 MW is
obtained by using the RBF neural network has been given in Table 3.1.
Table 3.1
Performance Indices and Contingency Ranking using RBF ANN for 5-BUS SYSTEM
Contingency PIP PIV Ranking
Number (Line Outage No.)
1 0.2908 3.7433 1
2 0.3755 0.2773 7
3 0.3302 0.6739 5
4 0.3926 0.7281 4
5 0.4149 0.3945 6
6 0.3021 0.9203 2
7 0.3047 0.8791 3
Table 3.2 shows the comparative results of the active power performance index PIP
obtained using fast decoupled load flow solution and using RBF neural network. Fig 3.6
shows the graphical representation of the closeness of the results obtained using FLDF and
54
RBF-ANN. It is found that the network fits the desired data well for seven hidden neuron
in the hidden layer and for a spread value of 𝛔=2.5.
Table 3.2
Active Power Performance Index using FDLF& RBF ANN for 5-BUS SYSTEM
Contingency PIP by PIP by Error
FDLF RBF ANN
Number (Line Outage No.)
0.8
PIp by FDLF
0.7 PIp by RBF-ANN
0.6
0.5
PIp
0.4
0.3
0.2
0.1
0
1 2 3 4 5 6 7
Contingency Number
Table 3.3 shows the comparative results of the reactive power performance index
PIV obtained using fast decoupled load flow solution and using RBF neural network. Fig
3.7 shows the graphical representation of the closeness of the results obtained using FLDF
and RBF-ANN.
55
Table 3.3
Reactive Power Performance Index using FDLF& RBF ANN for 5-BUS SYSTEM
4
PIv using FDLF
3.5 PIv using RBF-ANN
2.5
PIv
1.5
0.5
0
1 2 3 4 5 6 7
Contingency Number
56
3.6.2 Case II: 14- Bus Test System
The bus data and line data specification for the fourteen bus test system has been
given in APPENDIX B. The results of active power performance index PIP and reactive
power performance indices PIV obtained by using the RBF neural network for a loading
level of 10% above the base case i.e. at 2849 MW are given in Table 3.4.
Table 3.4
Performance Indices & Contingency Ranking using RBF-ANN for 14-Bus System
20 1.0500 2.2493 17
57
Table 3.5 shows the comparative results of the active power performance index PIP
obtained using fast decoupled load flow solution and using RBF neural network. Fig 3.8
shows the graphical representation of the closeness of the results obtained using FLDF and
RBF-ANN. It is found that the network fits the desired data well for fifteen hidden neuron
in the hidden layer and for a spread value of 𝛔=10.
Table 3.5
Active Power Performance Index using FDLF& RBF ANN for 14-BUS SYSTEM
Contingency PIP by PIP by Error
Number (Line Outage No.) FDLF RBF ANN
1 1.1693 1.2028 -0.0335
58
3
PIp by FDLF
PIp by RBF-ANN
2.5
PIp by FLDF & RBF-ANN
1.5
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Contingency Number
Table 3.6 shows the comparative results of the reactive power performance index
PIV obtained using fast decoupled load flow solution and using RBF neural network. Fig
3.9 shows the graphical representation of the closeness of the results obtained using FLDF
and RBF-ANN.
Table 3.6
Reactive Power Performance Index using FDLF& RBF ANN for 14-BUS SYSTEM
Contingency PIV by PIV by Error
Number (Line Outage No.) FDLF RBF ANN
1 7.2973 10.2083 -2.9110
59
9 10.5730 12.3597 -1.7867
14
13 PIv by FDLF
12 PIv by RBF-ANN
11
10
PIv by FDLF & RBF-ANN
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Contingency Number
60
Comparison of computation time: The computation time taken for contingency
selection by FDLF and RBF-ANN method for the two bus systems have been detailed in
Table 3.7.
Table 3.7
Computation time by FDLF& RBF- ANN
Bus System By FDLF By RBF
5-Bus 4.9 Sec 0.007 Sec
14-Bus 17.4 Sec 0.04 Sec
From the above table it can easily inferred that the contingency selection process
by the use of RBF-ANN is much faster as compared to the FDLF method. The same
inference has been illustrated in Fig. 3.10.
By FDLF By RBF-ANN
18 0.05
16 0.045
0.04
14
0.035
12
Time in Seconds
Time in Seconds
0.03
10
0.025
8
0.02
6
0.015
4
0.01
2 0.005
0 0
5 Bus 14 Bus 5 Bus 14 Bus
61
3.7 CONCLUDING REMARKS
The algorithm for performing the contingency analysis for five bus and fourteen
bus test system have been implemented using Radial Basis Function neural network, the
main focus has been to perform the fast contingency selection for the possible line
contingencies by calculating the two types of performance indices namely PIP and PIV. It
has been observed that when correct number of hidden neurons and bias centres are
chosen for the network the results obtained using RBF-ANN are very much close to that
which has been obtained using FDLF. But the neural network has the ability to perform
the contingency selection for any loading and generating conditions once it gets correctly
trained. Thus the algorithm provides a much faster and accurate solution for contingency
analysis. The prediction of performance indices is instantaneous when calculated though
RBF-ANN and hence it can put to use for online applications in power system.
62
CHAPTER-4
4.1 CONCLUSIONS
In this work, the contingency selection and ranking which are important for
contingency analysis have been done by evaluating two important performance indices
namely; active and reactive power performance index (PIP & PIV). These indices were
calculated for various test bus systems using the Fast Decoupled Load Flow (FDLF)
algorithm and also by using Radial Basis Function (RBF) Neural Network in MATLAB
environment. The study has been carried out for the three test systems namely 5-Bus,
IEEE-14 Bus and IEEE-30 Bus, the satisfactory performance has been obtained for 5-Bus
and 14-Bus system using RBF. The following conclusions are drawn:
The severity of a single line outage is accurately indicated by the numerical values
of PIP and PIV respectively.
The indices are predicted in off line manner for a single loading condition by
FDLF. The calculation of these indices using FDLF algorithm proves to be time
consuming.
The contingency selection by RBF-ANN proves to be efficient in terms of
accuracy and time. It has the ability to calculate the performance indices following
a contingency for any loading case once it is effectively trained.
The training for large systems like 30-Bus system and the structure of RBF-ANN
like, neurons in hidden layer need to be investigated further.
To perform the contingency analysis and the contingency selection considering a
multiple line or equipment failures.
Implement a hardware model for the neural network so that it can be used for
online applications in power system contingency analysis.
63
REFERENCES:
[1] Wood A.J and Wollenberg B.F., “Power generation, operation and control”, John
Wiley & Sons Inc., 1996.
[2] Stott B, Alsac O and Monticelli A.J, “Security Analysis and Optimization”, Proc.
IEEE, vol. 75,No. 12, pp. 1623-1644,Dec 1987.
[3] Lee C.Y and Chen N, “Distribution factors and reactive power flow in transmission
line and transformer outage studies”, IEEE Transactions on Power systems, Vol. 7,No.
1,pp. 194-200, February 1992.
[4] Singh S.N and Srivastava S.C, “Improved voltage and reactive distribution factor for
outage studies”, IEEE Transactions on Power systems, Vol. 12, No.3, pp.1085-1093,
August 1997.
[5] Peterson N.M, Tinney W.F and Bree D.W, “Iterative linear AC power flow solution
for fast approximate outage studies”, IEEE Transactions on Power Apparatus and
Systems, Vol. PAS-91, No. 5, pp. 2048-2058, October 1972.
[6] Brandwjn V and Lauby M.G, “Complete bounding method for a.c contingency
screening”, IEEE Transactions on Power systems, Vol. 4, No. 2, pp. 724-729, May 1989.
[8] Zaborzky J, Whang K.W and Prasad K, “Fast contingency evaluation using concentric
relaxation”, IEEE Transactions on Power systems, Vol. PAS-99, No. 1, pp. 28-36,
February 1980.
[9] Stott B and Alsac O, “Fast decoupled load flow”, IEEE Transactions on Power
Apparatus and Systems, Vol. PAS-91, No. 5, pp. 859-869, May 1974.
[10] Ejebe G.C and Wollenberg B.F, “Automatic Contingency Selection”, IEEE
Transactions on Power Apparatus and Systems, Vol. PAS-98, No. 1, pp. 97-109, January
1979.
64
[11] Warwick K., Ekwue A. and Aggrawal,R., “Artificial Intelligence Techniques in
Power systems”, The Institution of Eletrical Engineers, London,UK,1997.
[12] Pang C.K., Prabhakara F.S., El-Abiad A.H. and Koivo A.J., “Security evaluation in
power systems using pattern recognition”, IEEE Transaction on Power System Apparatus
and Systems,PAS-93,(2),pp.969-976 ,May/June 1974.
[13] Dillon T.S., Morsztyn K. and Plava K., “Short term load forecasting using adaptive
pattern recognition and self-organising techniques” Proc of the Fifth Power Sytem
Computation Conference, paper 2.413, pp.1-16, Cambridge 1975.
[14] Saito O., Koizumi K., Udo M., Sato M., Mukae H. and Tsuji T., “Security monitoring
systems including fast transient stability studies”, IEEE Trans. Power Apparatus and
Systems, PAS(5),pp.1789-1805, September/October 1975.
[15] Sobajic D.J and Pao,Y.H.,“Artificial neural net based dynamic security assessment of
electric power systems”, IEEE PES 1988.
[16] Aggoune M., El-Sharkawi M.A., Park D.C., Damborg M.J. and Marks II R.J.,
“Preliminary results on using artificial neural networks for security assessment”, IEEE
Proceedings of 1989 PICA, Seattle,Washington,USA, pp. 252-258, May 1989.
[17] Rajasekaran S. and Vijayalakshmi G.A., “Neural Networks, Fuzzy logic and Genetic
Algorithm Synthesis and applications”, (PHI Learning Private Limited, 2010).
[18] Scala M.La, Trovato M and Torelli F, “A neural network based method for voltage
security monitoring”, IEEE Transactions on Power Systems, Vol. 11, No. 3, August 1996.
[19] Zhou D.Q and Annakage U.D, “Online Monitoring of voltage stability margin using
artificial neural metwork”, IEEE Transactions on Power Systems, Vol. 25, No. 3, pp.
1566-1574, August 2010.
[20] Hiebur D, “Artificial neural networks in the power industry, survey and application”,
Neural Network World, Vol. 5(6), pp. 551-564, 1995
65
[21] Niebur D and Germond A.J, “Power system static security assessment using the
Kohenen neural network classifier”, IEEE Transactions on Power Systems, Vol. 7, No. 2,
pp. 865-872, May 1992.
[22] Boudour M. and Hellal A., “Combined use of unsupervised and supervised learning
for Large Scale power system Static Security Mapping”, IEEE conference on Power
Systems, pp.1321-1327, 2004.
[23] Lo K.L., Peng L.J., Macqueen J.F., Ekwue A.O and Cheng,D.T.Y, “Fast Real Power
Contingency Ranking using a counter propagation Network”, IEEE Transactions on
Power Systems,Vol. 13, No. 4, pp.1259-1264, November 1998.
[24] Pandit M., Srivastava L., Singh V., Sharma J, “ Coherency-based fast voltage
contingency ranking employing counterpropagation neural network”, Science Direct
Engineering applications of Artificial Intelligence.,Vol 20, pp.1133-1143 ,2007.
[25] Sidhu T.S and Cui L., “Contingency Screening for Steady State Security Analysis By
using FFT and Artificial Neural Networks”,IEEE Transactions on Power system,Vol.
15,No. 1, pp.421-426, February 2000.
[26] Chow J.C, Fischl R, Kam M, Yan H.H and Ricciardi S, “An improved Hopfield
model for system contingency classification”, Proceedings of the IEEE International
Symposium on circuits and systems, Vol. 4, pp. 2925-2928, 1990.
[28] Simon Hykins, “Neural Networks A comprehensive Foundation”, (Prentice Hall Inc.,
New Jersey, 1990)
[29] Refaee J.A, Mohandes,M and Maghrabi H, “Radial Basis Function for Contingency
Analysis of Bulk Power Systems”, IEEE Transactions on Power Systems,Vol. 14, No. 2,
pp.772-778 ,May 1999.
[30] Chicco G., Napoli,R and Piglione,F, “ Neural networks for fast voltage prediction in
power systems”,IEEE Porto Power Tech Conference,September 2001.
66
[31] Jain T., Srivastava S.N. and Jain A.,: “Parallel Radial Basis Function Neural Network
Based Fast Voltage Estimation for Contingency Analysis”, IEEE International Confrence
on Electric Utility Deregulation,Restructuring and Power Technologies
(DPRPT2004),,Hong Kong, pp.780-784, April 2004.
[32] Yuan Y.Hsu, Kua H.C, “Fuzzy set based contingency ranking”, IEEE Transactions
on Power Systems, Vol. 7, No. 3, pp. 1189-1196, August 1992
[33] Lo K.L and Abdelal A.K.I, “Fuzzy logic based contingency analysis”, DPRT 2000
International conference on Electric Utility Deregulation and Restructuring and Power
Technologies Proceedings, Cat. No. 00EX82, pp. 499-504.
[34] Nims J.W, El-Keibb A.A and Smith R.E, “Contingency ranking for voltage stability
using genetic algorithm”, Vol. 43, Issue. 1, pp. 69-76, October 1997.
[35] Ozdemir A, Lim J.Y and Singh C, “Contingency screening for steady state security
analysis using genetic algorithms”, IEEE Power Engineering Society Summer Meeting,
Vol. 3, pp. 1142-1147, 2002.
[36] Uma Rao.K., “Computer Techniques and Models in Power System”, (I.K.
International Publishing House Pvt. Ltd, 2007).
[37] Pai M.A., “Computer techniques in power system analysis”, (Tata McGraw Hill Ltd,
1998).
[38] Sadat H., “Power System analysis”, (Tata McGraw Hill Ltd, 2002)
67
APPENDIX-A
The bus data and line for the 5 bus test system has been given in Table A.1 and A.2
respectively. The following conventions were used for all the test bus systems; Base MVA
= 100; Coding used for buses: 0-Load Bus, 1-Slack Bus, 2-PV Bus.
Table A.1
Bus Data of 5-Bus System
Bus Bus Voltage Angle Load Load Gen. Gen. Gen. Gen. Injec.
No. code Mag. Degree MW MVAR MW MVAR Qmin Qmax MVAR
(pu) (pu) (pu) (pu)
1 1 1.06 0.0 0.00 0.00 0.0 0.0 0.0 0.0 0.0
2 0 1.0 0.0 0.20 0.10 0.4 0.3 0.0 0.0 0.0
3 0 1.0 0.0 0.45 0.15 0.0 0.0 0.0 0.0 0.0
4 0 1.0 0.0 0.40 0.05 0.0 0.0 0.0 0.0 0.0
5 0 1.0 0.0 0.60 0.15 0.0 0.0 0.0 0.0 0.0
Table A.2
Line Data of 5-Bus System
Start Bus End Bus R (pu) X (pu) ½ B (pu) Tap Set
value
1 2 0.0200 0.0600 0.0300 1
1 3 0.0800 0.2400 0.0250 1
2 3 0.0600 0.1800 0.0200 1
2 4 0.0600 0.1800 0.0200 1
2 5 0.0400 0.1200 0.0150 1
3 4 0.0100 0.0300 0.0100 1
4 5 0.0800 0.2400 0.0250 1
68
APPENDIX-B
The bus data and line data of IEEE-14 bus system has been given in Table B.1 and B.2
respectively.
Table B.1
Bus Data of IEEE-14 Bus System
Bus Bus Voltage Angle Load Load Gen. Gen. Gen. Gen. Injec.
No. code Mag. Degree MW MVAR MW MVAR Qmin. Qmax. MVAR
1 2 1.06 0.0 0.0 0.0 232.0 0.0 -10.0 10.0 0
2 1 1.05 0.0 21.7 12.7 40.0 -42.4 -0.4 0.5 0
3 2 1.04 0.0 94.0 19.0 0.0 0.0 0.0 0.4 0
4 0 1.0 0.0 47.8 0.00 0.0 0.0 0.0 0.0 0
5 0 1.0 0.0 7.6 1.60 0.0 0.0 0.0 0.0 0
6 2 1.08 0.0 11.2 7.5 0.0 0.0 -0.06 0.24 0
7 0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0
8 2 1.071 0.0 0.0 0.0 0.0 0.0 -0.06 0.24 0
9 0 1.0 0.0 29.5 16.6 0.0 0.0 0.0 0.0 0
10 0 1.0 0.0 9.00 5.80 0.0 0.0 0.0 0.0 0
11 0 1.0 0.0 3.50 1.80 0.0 0.0 0.0 0.0 0
12 0 1.0 0.0 6.10 1.60 0.0 0.0 0.0 0.0 0
13 0 1.0 0.0 13.50 5.80 0.0 0.0 0.0 0.0 0
14 0 1.0 0.0 14.90 5.00 0.0 0.0 0.0 0.0 0
69
Table B.2
Line Data of IEEE-14 Bus System
70
APPENDIX-C
The bus data and line data has been given in Table C.1 and C.2 respectively.
Table C.1
Bus Data of IEEE-30 Bus System
Bus Bus Voltage Angle Load Load Gen. Gen. Gen. Gen. Injec.
No. code Mag. Degree MW MVAR MW MVAR Qmin. Qmax. MVAR
1 1 1.06 0 0.0 0.0 0.0 0.0 0 0 0
2 2 1.043 0 21.70 12.7 40.0 0.0 -40 50 0
3 0 1.0 0 2.4 1.2 0.0 0.0 0 0 0
4 0 1.06 0 7.6 1.6 0.0 0.0 0 0 0
5 2 1.01 0 94.2 19.0 0.0 0.0 -40 40 0
6 0 1.0 0 0.0 0.0 0.0 0.0 0 0 0
7 0 1.0 0 22.8 10.9 0.0 0.0 0 0 0
8 2 1.01 0 30.0 30.0 0.0 0.0 -10 40 0
9 0 1.0 0 0.0 0.0 0.0 0.0 0 0 0
10 0 1.0 0 5.8 2.0 0.0 0.0 0 0 19
11 2 1.082 0 0.0 0.0 0.0 0.0 -6 24 0
12 0 1.0 0 11.2 7.5 0.0 0.0 0 0 0
13 2 1.071 0 0.0 0.0 0.0 0.0 -6 24 0
14 0 1.0 0 6.2 1.6 0.0 0.0 0.0 0.0 0
15 0 1.0 0 8.2 2.5 0.0 0.0 0.0 0.0 0
16 0 1.0 0 3.5 1.8 0.0 0.0 0.0 0.0 0
17 0 1.0 0 9.0 5.8 0.0 0.0 0.0 0.0 0
18 0 1.0 0 3.2 0.9 0.0 0.0 0.0 0.0 0
19 0 1.0 0 9.5 3.4 0.0 0.0 0.0 0.0 0
20 0 1.0 0 2.2 0.7 0.0 0.0 0.0 0.0 0
21 0 1.0 0 17.5 11.2 0.0 0.0 0.0 0.0 0
22 0 1.0 0 0.0 0.0 0.0 0.0 0.0 0.0 0
23 0 1.0 0 3.2 1.6 0.0 0.0 0.0 0.0 0
24 0 1.0 0 8.7 6.7 0.0 0.0 0.0 0.0 4.3
25 0 1.0 0 0.0 0.0 0.0 0.0 0.0 0.0 0
26 0 1.0 0 3.5 2.3 0.0 0.0 0.0 0.0 0
71
27 0 1.0 0 0.0 0.0 0.0 0.0 0.0 0.0 0
28 0 1.0 0 0.0 0.0 0.0 0.0 0.0 0.0 0
29 0 1.0 0 2.4 0.9 0.0 0.0 0.0 0.0 0
30 0 1.0 0 10.6 1.9 0.0 0.0 0.0 0.0 0
Table C.2
Line Data of IEEE-30 Bus System
72
18 19 0.0639 0.1292 0.0 1
19 20 0.0340 0.0680 0.0 1
10 20 0.0936 0.2090 0.0 1
10 17 0.0324 0.0845 0.0 1
10 21 0.0348 0.0749 0.0 1
10 22 0.0727 0.1499 0.0 1
21 22 0.0116 0.0236 0.0 1
15 23 0.1000 0.2020 0.0 1
22 24 0.1150 0.1790 0.0 1
23 24 0.1320 0.2700 0.0 1
24 25 0.1885 0.3292 0.0 1
25 26 0.2544 0.3800 0.0 1
25 27 0.1093 0.2087 0.0 1
28 27 0.0000 0.3960 0.0 0.968
27 29 0.2198 0.4153 0.0 1
27 30 0.3202 0.6027 0.0 1
29 30 0.2399 0.4533 0.0 1
8 28 0.0636 0.2000 0.0214 1
6 28 0.069 0.0599 0.065 1
73